Abstract
Objective
To compare possible differences in the proportion of medication errors associated with high-risk medications that were avoided by the use of automated infusion device (AID) technology in pediatric and adult intensive care unit (ICU) patients. A secondary purpose was to investigate the number of serious adverse drug events (ADEs) identified by root-cause analyses (RCA).
Method
The study included pediatric and adult patients receiving high-risk medications by continuous infusion in an academic medical center with mixed medical-surgical ICUs. A retrospective evaluation of 1 year's data collected prospectively in an AID database was used to compare the proportion of medication errors avoided based on reprogramming events (2.5 times limit as a low threshold) and overrides (10 times limit as high). Information obtained from RCAs was used to compare the proportion of serious ADEs that occurred during the 5-year periods before and after AID implementation.
Results
The pediatric population was 1.68 times (95% confidence interval [CI], 1.18 to 2.38) more likely to require a reprogramming event than the adult acute care population for all high-risk medications combined. Significantly more reprogramming events occurred in the pediatric patients with potassium (relative risk [RR], 2.77; 95% CI, 1.15 to 6.68) and insulin (RR, 2.73; 95% CI, 1.15 to 6.45) infusions. Additionally, there were more overrides in the pediatric compared to the adult population for the high-risk medications (RR, 1.82; 95% CI, 1.32 to 2.53). The number of serious adverse or sentinel events as identified in RCAs decreased from six before (four deemed preventable by AID technology) to three (zero preventable) after AID implementation.
Conclusions
This study demonstrates that AID technology when properly used leads to reductions in medication errors and possibly serious ADEs in critically ill patients receiving high-risk medications. The technology appears to be particularly beneficial in pediatric patients with weight-based dosing strategies. However, the potential for clinicians to override the alerts remains a concern.
The medication-use process is prone to medication errors and associated adverse drug events (ADEs) because health system personnel are required to perform multiple functions in cooperation with each other and machines. 1 System redundancy helps to limit preventable ADEs in earlier phases but is less likely to be available for circumventing ADEs originating in the terminal medication administration phase. 2 Of particular concern are administration-associated ADEs involving high-risk medications such as heparin and insulin commonly given by the intravenous (IV) route. The term high-risk or high-alert refers to medications that are more likely to result in severe ADEs when used in error.3–6
New safety technologies have emerged aimed at reducing the number and severity of preventable ADEs associated with the administration phase of the medication use process. One example, automated infusion devices (AIDs), also referred to as smart pumps, for the administration of IV medications have been available for decades. However, some of the earlier models of AIDs were prone to administration phase ADEs due to keypad data entry mistakes as well as slips or memory lapses while entering data. 7 Newer AID technology consists of a pump and a computer processor that combine drug delivery, monitoring, and data management functions. The multifunctional safety software is designed to reduce medication errors and increase patient safety by enabling safer medication administration and monitoring of IV therapies.
The AID software allows each institution to create a dataset that contains minimum and maximum parameters (ie, dose, duration, concentration, etc) for each medication. The minimum parameter is important from an efficacy standpoint, and the maximum parameter is important from a toxicity standpoint. A dataset contains multiple profiles that allow the same drug to have different limits depending on various patient parameters. The software limits are set by hospital personnel. Once the dataset has been created, if a clinician enters a dose or rate that is outside the limit(s) and presses the start button, an alert appears and the infusion will not begin until the clinician has responded to the alert.
There are a number of review articles and commentaries that discuss the potential benefits and cost-effectiveness of AID technology, but evidence from investigations supporting its use is limited.2,3,8–12 The studies that are available tend to fall into two categories, those that assess AID technology as part of a multifaceted approach for reducing medication error and those that investigate overall medication errors before and after AID implementation.9–12 Only one of the studies focused on patients in an intensive care unit (ICU; ie, cardiac surgical ICU and step-down units), but that study was focused on adults and had a major limitation in that nurses could easily bypass the AID limits. 9 With the latter exception and one retrospective study involving anticoagulants, 12 none of the previously mentioned studies examined specific high-risk medications and/or electrolytes in the ICU setting. In particular, there is a need for research comparing the benefits of AID technology in pediatric and adult populations, since the majority of medication dosing in children is weight-based and the additional calculation and lack of a “normal” dose per unit of time (eg, mg/h or mg/d) as in adults introduces additional sources of error.
The primary purpose of this two-part retrospective study was to compare possible differences in the proportion of medication errors associated with high-risk medications that were avoided by the use of AID technology in pediatric and adult ICU patients. A secondary purpose was to investigate the number of serious adverse or sentinel events associated with medications administered by continuous infusion as identified by root-cause analyses (RCA) that were conducted before and after AID implementation.
Materials and Methods
Study Site
This retrospective study was conducted at a 400-bed tertiary care academic medical center in Tucson, Arizona. At the time of this study, the hospital had a 16-bed pediatric ICU and approximately 16-bed medical, surgery/trauma, and cardiothoracic adult ICUs. The center implemented AID technology (Alaris Medical Systems, Cardinal Health, Inc., Dublin, Ohio, USA) in 2003 as its sole IV infusion pump. Several profiles are available with this technology, including weight-based profiles for pediatrics starting at 0 to 5 kg and three different adult profiles for ICU, non-ICU, and oncology patients. All IV medications (continuous, intermittent, and maintenance) are administered with the infusion pumps. All new nurses undergo a 2-hour training class before using AIDs, and “super users” will periodically round to ensure that all pumps are being used appropriately. The AID database has distinct profiles with minimum and maximum dosing limits for pediatric and adult patients. Limits for the pumps are entered by a pharmacist with expertise in information technology after receiving recommendations from pharmacists, nurses, and physicians at the institution. These clinicians give recommendations with respect to their areas of expertise (eg, critical care clinicians for ICU medications). Medications to be evaluated were identified from two sources based on their potential to cause patient harm: (1) high-alert medications compiled by the Institute for Safe Medication Practices that are likely to be used in an ICU setting, 5 and (2) medications identified by the investigators as high-risk based on their previous research involving pediatric and adult ICU settings.7,13 The study was performed with the approval of the Institutional Review Board of the University of Arizona.
Design
This was a two-part retrospective study involving data obtained from an AID database and RCAs. For the first part of this study, information on high-risk medications obtained from the AID database was used to compare the proportion of medication errors avoided by the use of AID technology in pediatric and adult patients in mixed medical-surgical ICUs. We investigated this endpoint by using a low threshold of reprogramming events (as defined by infusions exceeding 2.5 times the minimum or maximum limits) and a high threshold of overrides (exceeding 10 times the predefined limits). The proportion of reprogramming events was considered as a low threshold for error, because some of these events may have been due to unnecessarily restrictive limit placement by our clinicians. The 10-fold override limit was considered a high threshold; an override exceeding this amount would be unlikely to be justifiable as a legitimate dose in most patient care situations. The 2.5- and 10-fold limits that were chosen are consistent with the low and high indices utilized in previous investigations of AID technology.9–11 For the second part of this study, information on high-risk medications (administered by continuous infusion) obtained from RCAs conducted on serious adverse or sentinel events was used to compare the proportion of serious ADEs that occurred during the 5-year period before and after AID implementation.
Definitions
An alert is generated whenever a clinician enters a value that exceeds the predefined minimum or maximum limits of the medication as programmed in the AID software. The clinician must address the alert by either reprogramming or overriding. For this study, a reprogramming event occurred when the clinician received an alert that the hard or soft limits were exceeded, realized the mistake or programming error, and reentered the data. If the clinician received an alert and reprogrammed the pump with new values that fell within the defined limits, it was considered an avoided medication error. If the clinician received an alert and proceeded to infuse the medication ignoring the alert or reprogrammed the pump with the same values or values that were still outside the limits, it was an override. For this study, if the override exceeded 10 times the predefined limit, it was considered a medication error that was not avoided.
Data Collection
When an alert is generated on the infusion device, the infusion device wirelessly sends the data to a central server where they can be retrieved using an internal reporting tool. For the first part of this study, data from the AID database must have been logged into the server during the time of January 1 through December 31, 2007. Information was retrieved from the AID database by a pharmacist who specialized in information technology. Specific parameters were set to retrieve alerts pertaining to IV infusions of heparin, insulin, morphine, hydromorphone, fentanyl, vasopressin, nicardipine, potassium, magnesium, and calcium. Although all alerts involving adult ICU, adult acute care, and pediatric profiles were included, it was assumed that the vast majority of alerts associated with the AID technology occurred in pediatric and adult ICUs given the nature of the high-risk medications being studied. This assumption is supported by hospital policy that restricts the use of the high-risk medications listed in the pediatric profiles, with the exception of the electrolytes, to the pediatric ICU. The only alerts counted for the purposes of this study were dose, rate, and concentration of the high-risk medications. The total number of infusions administered was determined by retrieving the number of IV solutions for the specific medications entered into the pharmacy's order entry system, Centricity 8.00 (GE Healthcare, Barrington, Illinois, USA). Infusions that contained heparin at concentrations not considered therapeutic and mixed electrolyte infusions and maintenance infusions that contained minimal amounts of electrolytes were excluded. Also excluded were any infusion administered by a patient-controlled analgesia pump and any medication that was administered IV push or without the use of AID technology (eg, pressure infusions, free-flow infusions).
For the second part of this study, all RCAs conducted on serious adverse or sentinel events evaluated by the Quality Review Committee (QRC) within the timeframe of March 26, 1998 to March 26, 2008, were initially reviewed by an intensivist who was a member of the QRC. This time frame included a 5-year period prior to AID implementation and a 5-year period after AID implementation. All events related to IV infusions were collected, de-identified, and subsequently reviewed by another intensivist and a pharmacist specializing in critical care to ensure that each event was potentially preventable by AID technology.
Statistical Analysis
The proportions of medication errors associated with high-risk medications administered to pediatric and adult patients that were avoided by the use of AID technology were compared using Fisher exact tests with data reported as relative risk (RR) and 95% confidence intervals (CI). Significance for all comparisons was defined as P < 0.05. All statistical programming was performed using Intercooled Stata 7.0 (StataCorp LP, College Station, Texas, USA).
Results
Reprogramming Events
A total of 60,419 infusions involving the specified medications were administered during the 1-year period of data collection (see Table 1). Potassium chloride was the most commonly administered high-risk medication, with 27,153 infusions combined between the two groups. A total of 261 infusions, 225 in the adult and 36 in the pediatric groups, generated an alert in which the final outcome resulted in a reprogramming event when the limit was exceeded by 2.5 times or greater. Fentanyl incurred the greatest number of reprogramming events with 66 events occurring in the adult and 16 in the pediatric groups. For heparin, of the four alerts generated in children, three were associated with the 0 to 5 kg profile that has the nurse enter the dose in units/kg/h, whereas all the other profiles have the dose entered in units/h. Overall, the pediatric population was 1.68 times (95% CI, 1.18 to 2.38) more likely to require a reprogramming event than the adult acute care population for all high-risk medications combined. With respect to specific medications, significantly more reprogramming events occurred in the pediatric patients with potassium (RR, 2.77; 95 CI, 1.15 to 6.68) and insulin (RR, 2.73; 95% CI, 1.15 to 6.45) infusions (see Table 1). For potassium, all of the alerts in adults were related to dose, while the alerts in children were a function of dose (66.7%) and concentration (33.4%). For insulin, 94% of the alerts in adults and 16.7% in children were related to continuous infusions; the other 88.3% of the alerts in children were related to bolus doses. All five of the pediatric alerts below the minimum in Table 1 occurred in the 0 to 5 kg profile and were below the 0.05 units/min limit of the device.
Number of reprogramming events 2.5 times above or below the predefined limits for high-risk medications in adult and pediatric intensive care units a
Note: CI = confidence interval; RR = relative risk.
Significance for all comparisons was defined as P < 0.05.
Max defined as > 2.5 × above limit.
Min defined as < 2.5 × below limit.
RR cannot be calculated if there is no data value in a cell.
Significantly more reprogramming events in pediatric population.
Overrides
Table 2 shows the number of alerts when the maximum limit was exceeded by 10 times or greater, but the clinician generated an override. The adult population showed an increased number of overrides when compared to the number of reprogramming events. Overall, there were significantly more overrides in the pediatric compared to the adult population for the high-risk medications listed in Table 2 (RR, 1.82; 95% CI, 1.32 to 2.53); however, there were significantly fewer overrides in the pediatric versus adult patients on fentanyl (RR, 0.34; 95% CI, 0.17 to 0.70). All of the adult alerts leading to fentanyl overrides were from continuous infusion dose limits, whereas the pediatric alerts leading to overrides were spread between bolus dose (37.5%), concentration (12.5%), and continuous infusion dose (50%) limits. In looking at the fentanyl continuous dose alerts, 89 of the 91 alerts were from a rate of 999 mL/h being entered.
Number of overrides exceeding 10 times the predefined limits for high-risk medications in adult and pediatric intensive care units a
Note: CI = confidence interval; ICU = intensive care unit; RR = relative risk.
Significance for all comparisons was defined as P < 0.05.
RR cannot be calculated if there is no data value in a cell.
Significantly more overrides in the pediatric population.
Significantly more overrides in the adult population.
Table 3 provides some examples of reprogramming events that were discovered in this study.
Examples of reprogramming events and final outcome
There were nine serious ADEs involving infused medications that were the subject of RCAs for the 5-year periods before and after AID implementation. Of the six RCAs conducted prior to AID implementation, one involved the incorrect preparation of an electrolyte solution that likely would not have been prevented had the technology been available. The other five RCAs were medication administration errors that involved incorrect mode or site of delivery, incorrect rate of infusion, improper cessation of IV infusions, and wrong medication administered. Four of these five serious ADEs were deemed preventable had AID technology been available. The improper cessation event was deemed nonpreventable by the technology because the user deliberately turned off medication infusion pumps to complete another task.
Three RCAs were conducted after AID implementation, none of which were amenable to prevention with this technology. Two of these RCAs were attributed to incorrect ordering and one to improper medication preparation.
Discussion
Although adult patients in the hospital's ICUs received a greater number of infusions of high-risk medications, the pediatric population was 1.68 times more likely to require a reprogramming event. Significantly more reprogramming events occurred in the pediatric patients with potassium infusions (RR, 2.77; 95% CI, 1.15 to 6.68) and insulin (RR, 2.73; 95% CI, 1.15 to 6.45). Of greater concern is the increased number of 10-fold overrides in the pediatric compared to the adult population, since medication dosing this far beyond pump limits is unlikely to be justifiable based on specific patient characteristics. The overall benefits associated with AID technology were corroborated by the number of serious adverse or sentinel events as identified in RCAs that decreased from six before (four deemed preventable by AID technology) to three (zero preventable) after AID implementation.
The potential benefits in the pediatric patients with respect to reprogramming events was to some degree offset by the significantly (RR, 1.82; 95% CI, 1.32 to 2.53) greater number of overrides (ie, medication errors not avoided) that exceeded 10 times the upper limits programmed into the AID. One notable exception was the significantly (RR, 0.34; 95% CI, 0.17 to 0.70) greater number of overrides that occurred with fentanyl in adult patients. Reasons for these results are not easily defined and could involve multiple factors. In general, pediatric patients, especially the critically ill, are more prone to medication errors that involve high-risk medications. 14 One possible explanation for this finding is that weight-based dosing for pediatric patients is the norm and may lead to errors due to wrong choice of weight, multiplication errors such as weight-times-dose calculations, decimal point errors (eg, 10-fold errors), or difficulty in catching errors in the final calculated dose given the wide range of “normal” doses. Additionally, there is not a “normal” dose per unit of time (eg, mg/h or mg/d) that is often recognized when dosing adult patients, because the dosing requirements in children vary substantially depending on age and weight. Knowing this, clinicians in pediatric ICUs may pay closer attention to doses and infusion rates compared to those in adult ICUs, which would be consistent with the larger number of reprogramming noted in the pediatric patients in this study after an alert was generated.
The weight-based dosing explanation for the higher number of reprogramming events in the pediatric group is also consistent with the two medications, potassium and insulin, that accounted for the differences; these medications, with the possible exception of an initial bolus, are typically dosed and infused without regard to weight in adult patients in our institution. For example, with respect to potassium, the adult devices have a commonly used predefined concentration (eg, 10 or 20 mEq in 50 mL) in addition to an entry that allows a registered nurse to enter a nonstandard concentration, whereas the pediatric devices only have a nonstandard concentration entry.
Although the larger number of reprogramming events in the pediatric compared to the adult patients may be explained by possible differences in weight-based dosing and attention to doses and rates, it was interesting to note that the number of overrides was higher for the pediatric population as well, with the exception of fentanyl. Because 89 of the 91 alerts were from a rate of 999 mL/h being entered, this is probably a sign of the clinician either giving a bolus dose or flushing the line without using the correct function, bolus or flush, on the pump. A full explanation for the override findings remains to be elucidated, but this is clearly an area of concern given the high dose (ie, when the entered rate exceeded the maximum limit by 10 times or more) used to define an override in this study. In a previous study involving critically ill patients, serious medication errors were not reduced with AID implementation, which the investigators concluded was due to poor compliance issues such as overrides. 9 When looking at the differences, it is interesting to note that even though magnesium was never subject to a reprogramming event in the pediatric population, it was overridden on six occasions. Similarly, heparin was the subject of only four reprogramming events in the pediatric population but was overridden on eight occasions. This is in sharp contrast to a previous study involving smart infusion pumps that showed no override events when an alert was generated involving heparin. 12 The override finding is a subject of further investigation at the study institution, although one change has already been implemented to prevent egregious overrides such as those associated with fentanyl alerts. When possible, hard limits (ie, not amenable to change by the clinician) have been programmed into the AIDs to prevent overrides exceeding 10 times the low or high predefined limits.
The number of serious adverse or sentinel events conducted on RCAs associated with infused medications that were reported in the 5-year periods before and after AID implementation was small, four versus zero, respectively, but they support the benefits of this technology given that the ADEs were of sufficient severity to warrant time-consuming, in-depth analyses by hospital personnel.
This investigation extends the findings of previous investigations, 2,3,8,12 which have found reductions in medication errors and potential ADEs by AID technology in non-ICU settings, to high-risk medications in the ICU. However, further research is needed to document the cost-benefit or cost-effectiveness of this expensive technology that requires ongoing technical support in addition to start-up costs. The costs of the technology make it imperative that it is utilized to its fullest potential, which requires ongoing education and training of personnel to ensure consistent and proper use. For example, with the ability of a clinician to run an infusion in an open channel without using the safety software and the lack of complete integration with other medication software (eg, computerized prescriber order entry, barcoding), it is difficult to obtain the full benefits with respect to reductions in medication errors that may lead to ADEs. 9 Another important issue is the safety software's drug library. As is done in our hospital, hard and soft limits should be evaluated regularly and adjusted to meet the institution's needs. User fatigue can play a substantial role in whether or not the alert generated will result in any action by the clinician who sees it. If the limits are too stringent and alerts are generated constantly, the user will become accustomed to the alerts and meaningful alerts may be overridden. However, if the limits are set too loose, meaningful alerts will not be generated when they should be. Having these proper measures in place will help users to optimize the use of AID technology and hopefully reduce the possibility of human error.
In our institution, we are fortunate to have a pharmacist who is a dedicated information technologist. This pharmacist is not only responsible for the majority of AID profile programming but also co-chairs a committee with two nurses to ensure proper pump utilization. This committee meets on a monthly basis to evaluate data and recommend changes to the various profiles. Nevertheless, a number of the limits were reevaluated as a result of this study and a follow-up study is in progress that will assess the impact of AID technology on actual reported patient safety concerns such as ADEs.
There are several limitations to our study. As mentioned previously, the ability to run infusions without using the safety software was available and therefore we may not have had complete data on all of the medications evaluated. Institutional policy and expectations are that nurses will use the safety software whenever they are infusing a medication that is in the safety software. Reports show that on average over 80% of total infusions are given using the safety software. Given the retrospective nature of this investigation, it was not possible to obtain follow-up data for alerts that were overridden with continuation of the infusions. It is unknown how many of the reprogramming and override events resulted in actual patient harm, if any, or if actual ADEs were reduced with the use of the AID technology. The number of recorded sentinel events in the RCA analysis was small, but this is not surprising given their severity, the size of the hospital, and the investigation of only those events amenable or potentially amenable to AID technology. Finally, the human factors responsible for the recorded events (eg, staffing levels and user fatigue, lack of training or refresher training) could not be determined with this design, but this issue is deserving of further study.
Conclusion
This study demonstrates that AID technology when properly used leads to reductions in medication errors and possibly serious ADEs in critically ill patients receiving high-risk medications. The technology appears to be particularly beneficial in pediatric patients who are prescribed medications using weight-based dosing strategies. However, there are ongoing concerns related to the proper use of this expensive technology, particularly the potential for clinicians to override the alerts that are generated. More research is needed to identify human factors causing improper use of AID technology as well as economic investigations that identify those populations most likely to benefit from its implementation.
No financial support was used for this study. Dr Romero has received honoraria for meetings with Alaris.